EVE owns state
Project memory, rules, permissions, and execution stay in EVE, not inside a model's context window or chat history.
Personal Project · AI Orchestration
EVE
A persistent AI-assisted software development environment — a control plane that remembers project state, coordinates specialized AI coding agents, and turns natural-language intent into a planned, implemented, and auditable software project.
What EVE is
EVE (Extended Virtual Environment) started as a project dashboard and command launcher — a way to kick off builds, deployments, and scripts across my machines from one place. This document is the architecture for its next evolution: turning EVE into a system that understands my projects well enough to plan work, delegate it to the right AI coding agent, and remember what happened, days or months later, without me re-explaining everything from scratch.
The goal isn't a chatbot bolted onto a dashboard. It's a system that can take something like “let's build a volume controller app for Windows, with an iOS companion and physical knobs,” and turn that into a real project: a plan, a set of tasks, the right agent assigned to each one, and a durable memory of every decision made along the way.
EVE can also coordinate work across multiple isolated environments — for example a personal setup and a work setup — while keeping each one's machines, storage, and permissions scoped to itself instead of pooled into one global set of resources.
EVE should understand
- Me — preferences and working style
- My projects — history, architecture, current state
- My environments — which resources, machines, storage, and policies belong together
- Process — rules, dependencies, past decisions
- History — Git activity, agent activity, approvals
The core idea
EVE is the persistent system. The AI is interchangeable reasoning on top of it. That split is what keeps the whole thing from becoming a fragile wrapper around one model.
Models are swappable
The planning brain and coding agents — Claude, Codex, a local model — can change over time without rewriting EVE itself.
EVE is the broker
Agents don't launch each other directly. Every delegation flows through EVE, so it stays observable, permissioned, and safe.
Humans stay in control
Configurable autonomy levels and approval gates mean destructive or high-impact actions never happen silently.
How a request flows
A simplified view of the path from a user request to work actually getting done.
User
EVE
scoped to the current Environment
Planner
Specialized Agents
Machines
Git
Docker
Projects
Final product definition
EVE is a persistent AI-assisted software development environment and orchestration control plane that understands project state, remembers decisions, coordinates specialized AI workers, manages execution across local and remote machines, and allows me to create and evolve software through natural conversation instead of manually preprogramming every workflow.
Persistent
Model-agnostic
Conversational
Project-aware
Memory-driven
Tool-controlled
Machine-aware
Auditable
Modular
Safe by policy
Go deeper
The full specification covers all of it in detail — memory architecture, agent routing, Git isolation strategy, the planner loop, approval levels, and a complete worked example — across 54 sections with diagrams.
Architecture Specification
Rendered directly from the source Markdown document, unedited, including every diagram and callout.